MedPix-VQA Dataset

The MedPix-VQA dataset is a version of the data found at MEDPIX-ClinQA, specifically modified to address an image overlap issue that would result from directl splitting the original dataset. This overlap can lead to a model potentially seeing the same image during both training and validation, potentially leading to bias or data leakage.

Key Modifications:

• We have modified the dataset to ensure no image overlap between the training and validation sets. This was done by grouping the data based on image hashes and ensuring that all instances of an image (with different Q&A pairs) are either in the training set or in the validation set, but not in both.
• This ensures that the model's evaluation is done on truly unseen images, providing a more accurate assessment of its generalization performance.

Dataset Structure:

The dataset consists of the following columns:

• image_id: The ID for each image.
• mode: The mode or context for the image.
• case_id: The unique case identifier.
• question: A textual question related to the image.
• answer: The corresponding answer to the question.

Dataset Split:

• Training Set: 17420 examples (85%)
• Validation Set: 3080 examples (15%)

DatasetDict({
    train: Dataset({
        features: ['image_id', 'mode', 'case_id', 'question', 'answer'],
        num_rows: 17420
    })
    validation: Dataset({
        features: ['image_id', 'mode', 'case_id', 'question', 'answer'],
        num_rows: 3080
    })
})

Usage:

This dataset is designed for training and evaluating Visual Question Answering (VQA) models in the medical domain. It contains images from MedPix, a well-known medical image database, paired with questions and answers that focus on image interpretation.

📚 Citation

If you use this dataset, please cite the original authors:

@misc{siragusa2024medpix20comprehensivemultimodal,
  title={MedPix 2.0: A Comprehensive Multimodal Biomedical Dataset for Advanced AI Applications},
  author={Irene Siragusa and Salvatore Contino and Massimo La Ciura and Rosario Alicata and Roberto Pirrone},
  year={2024},
  eprint={2407.02994},
  archivePrefix={arXiv},
  primaryClass={cs.DB},
  url={https://arxiv.org/abs/2407.02994}
}